The Best Open-Source LLMs in 2026: 8 Open-Weight Models Ranked

Start with GLM-5.3 for a coding-focused evaluation, DeepSeek V4.1 Flash for a low-cost API, and gpt-oss-120b if your deployment budget is one 80GB GPU. Meta's Muse Spark Open adds a serious multimodal option, but downloadable weights do not make its commercial terms interchangeable with MIT or Apache 2.0.

The verdict

If you want an open model for long coding jobs, evaluate GLM-5.3 first, subject to its current license. If you want inexpensive hosted inference with permissively licensed weights available for a later move in-house, start with DeepSeek V4.1 Flash. If an Apache 2.0 checkpoint is a requirement, shortlist Qwen3-235B-A22B, gpt-oss-120b, or Mistral Small 4. Muse Spark Open belongs on the shortlist when your agent needs to interpret visual or audio material as well as text.

Since October 15, 2026, Meta has made Muse Spark Open's weights available, and NVIDIA has released Nemotron 3.5 Super VL. That expands the choice for multimodal agents beyond the older Llama comparison. The model cards, licenses and prices below were checked on October 16, 2026. Meta announcement, NVIDIA announcement.

An open-weight model is one whose trained parameters you can download and run yourself. That gives you another deployment option; it does not make serving free. These eight primary picks are ordered for this buying decision, balancing capabilities, license conditions and deployment practicality. They are an analysis of published evidence, not results from a hands-on comparison.

Context is the amount of material a model can handle in a request. It is not a promise that every host exposes that limit, or that every fact in a long document will be retrieved correctly. Check the exact checkpoint, host and serving configuration before sizing your application around it.

Why open weights stopped being a compromise

You can now choose downloadable models for coding, tool use and visual-document workflows without treating them as a separate class of toy systems. The useful question is whether a particular model completes your tasks within your latency, hardware and license constraints.

Benchmarks help narrow that evaluation, but the surrounding software matters. A harness is the program that gives the model tools, prompts and retry rules. Meta's Muse table, checked October 16, 2026, reports 76.4 on Terminal-Bench 2.1 using Terminus 2. Z.ai reports 88.2 for GLM-5.3 using Claude Code. Those are published results under different harnesses, not a controlled head-to-head test. Meta model card, Z.ai model card and methodology.

Open weight and open source are different claims. Download access tells you that you can obtain the parameters. A permissive weight license tells you what you may do with them. The Open Source Initiative's definition goes further, requiring appropriate access to code and training-data information as well as parameters. An MIT or Apache label on weights alone does not establish that the whole AI system meets that definition. Open Source AI Definition.

GLM-5.3 is the coding model to evaluate first

GLM-5.3, from Z.ai, is the first pick here for a team evaluating long coding and tool workflows. Its current documentation specifies text-only input, a 1M-token context window and always-on reasoning. That makes it a candidate for repository work, but not a standalone model for reading screenshots. A large context window can help keep code and instructions together; it does not eliminate the need to retrieve the right files. GLM-5.3 documentation.

For a coding agent, begin with the maker's recommended reasoning setting and compare completed patches, failed tool calls and elapsed time against your existing model. Z.ai's card lists deployment support including SGLang, vLLM and Transformers. These are inference engines, the software that loads the weights and serves requests. The choice of engine and quantization, which stores weights at lower precision, is part of your deployment specification. GLM-5.3 model card.

The license has changed. GLM-5.3 is not MIT: its license contains a security-review requirement for defined model-as-a-service businesses whose licensee-and-affiliate aggregate revenue exceeds $10 billion over any consecutive 12 months. The older GLM-5.2 card still identifies an MIT release. If standard permissive terms are your deciding constraint, evaluate that checkpoint or DeepSeek rather than carrying the old license label onto the new model. GLM-5.3 license, GLM-5.2 card.

DeepSeek V4.1 Flash is the budget API starting point

DeepSeek V4.1 Flash is the first hosted option to evaluate for extraction, summarization and other high-volume tasks where price matters. Its current card describes text and image input, a 1M-token context and MIT-licensed weights. It lists 552B backbone parameters, with conditional-memory parameters described separately, and 8B active parameters during prompt processing versus 16B during generation. A mixture-of-experts model activates only part of its weights for each token; that does not mean the rest of the weights disappear from the deployment. DeepSeek V4.1 Flash card.

The API name matters. DeepSeek's current pricing page maps deepseek-flash to V4.1 Flash; the legacy V4 Flash names now route to that model too. deepseek-v4-pro maps to V4-Pro-0813, the release that superseded the Pro preview. Flash supports vision on this API; Pro does not. Pin and record what your provider serves instead of assuming an old model name preserves an old model. Current API models, Pro-0813 card.

For a founder processing incoming documents, start with Flash and measure which cases fail. Move difficult cases to another model only when your evaluation demonstrates a benefit. The same rule applies to Pro: a larger model is an alternative to test, not an automatic upgrade for every request.

Kimi K3 is built for long multimodal agent work

Kimi K3, from Moonshot AI, is the newer model to evaluate for coding agents that also need visual understanding or broader knowledge work. Its card specifies 2.8T total parameters, 104B active parameters and a 1,048,576-token context window, with text, image and video understanding. That changes both the capability scope and the deployment scale from K2.7 Code. Kimi K3 model card.

For an internal development agent that must inspect a screenshot, change the application and check the result, a multimodal model can keep those steps within one model workflow. That is a reason to evaluate K3 alongside a text-focused coding model. It is not a reason to buy the hardware before testing the workflow through a host.

K3 uses the Kimi K3 License, not K2.7 Code's modified MIT terms. Its license includes a separate-agreement clause for qualifying model-as-a-service businesses, attribution conditions and explicit exceptions for defined internal use and access through official products or certified inference partners. Read those clauses against your intended deployment. Kimi K3 license.

The trade-off is deployment size and custom licensing. A team that needs a compact, permissively licensed text model should look elsewhere in this list. For the software that gives these models their tools and execution loop, see the best AI coding agents.

Qwen3 remains a permissive deployment option

Qwen3-235B-A22B, from Alibaba's Qwen team, remains a useful choice when you specifically want an Apache 2.0 checkpoint with multilingual support. Its card lists 235B total parameters, 22B active parameters, and support for more than 100 languages and dialects. It also supports switching between thinking and non-thinking modes. Qwen3-235B-A22B card, license.

Be precise about context: this original checkpoint has 32,768 native tokens, expandable to 131,072 with YaRN, a context-extension method. It is not the same release as every later model with a similar name. If you are building multilingual support-ticket triage, test the languages you actually receive and the context configuration you plan to serve.

It is also no longer true that Max-class Qwen weights are unavailable. Qwen3.8-2.4T-A95B is a published Max-class checkpoint, but it has its own custom license. Its hosted Qwen3.8-Max service adds features beyond the released language-model checkpoint. Keep this older Qwen3 entry on your shortlist for its specific license and deployment profile, not because it is the newest Qwen flagship. Qwen3.8 card, Qwen3.8-Max license.

MiniMax M3 brings multimodal long context

MiniMax M3 is an option when your work mixes text, images and video. Its current card describes approximately 428B total parameters, 23B active parameters and a 1M-token context window. It uses MiniMax Sparse Attention and supports enabled, adaptive and disabled reasoning modes. MiniMax M3 card.

For a product that summarizes a recording alongside related documents, the multimodal input is the reason to evaluate it. Use representative recordings and documents, then check whether the model cites the right evidence. Context capacity alone does not tell you how reliably it will find a brief event inside a long recording.

The commercial terms are more specific than a blanket requirement for a separate agreement. The license requires Built with MiniMax M3 attribution. It calls for prior written authorization when the relevant products or services generate more than $20 million in yearly revenue; otherwise it requires a one-time notice. Its prohibited-use terms also apply. Those conditions differ from Muse's model-as-a-service clause, even though both contain a revenue figure. MiniMax M3 license.

Choose M3 when its multimodal workflow earns its place and those terms fit. If avoiding custom license conditions is the priority, start with the Apache entries instead.

gpt-oss is the one that runs on your own GPU

gpt-oss, OpenAI's open-weight family, is the practical starting point when you need local text reasoning without a large cluster. The released MXFP4 version of gpt-oss-120b has 117B total parameters and 5.1B active parameters and fits on one 80GB GPU. The smaller gpt-oss-20b runs within 16GB of memory. Both use Apache 2.0. OpenAI's model card.

This is a deployment recommendation, not a compliance guarantee. Your serving process, logs, retrieval system and connected tools determine where data goes. Size the complete application, including working memory for concurrent requests, rather than treating the model's fit claim as a throughput promise.

The model supports configurable reasoning and tool calls, but your application must provide the tools and execute them. It also requires the harmony message format. The 120b model's documented context is 131,072 tokens, and it accepts text rather than images, audio or video. Model card, OpenAI model documentation.

For an NVIDIA deployment that needs vision, also evaluate Nemotron 3.5 Super VL as an alternative outside the eight ranked picks. NVIDIA's card specifies 120B total / 12B active parameters, up to 1M context, text/image/video input and OpenMDW-1.1 licensing; it explicitly allows commercial and non-commercial use. It lists vLLM, SGLang and TensorRT-LLM support. That makes it relevant to document-reading agents, but does not establish that your full workload fits the same hardware configuration as gpt-oss. NVIDIA model card.

Mistral is the Apache-licensed multimodal option

Mistral Small 4 is the Mistral checkpoint to evaluate here for a text-and-image agent under Apache 2.0. Its card specifies 119B total parameters, 6.5B active parameters and a 256K-token context, with instruction following, reasoning, function calling and JSON output. Mistral Small 4 card.

For a business processing scanned forms, that gives you a permissively licensed visual option without treating a text-only model as an image reader. Its card supports changing reasoning effort per request, so ordinary extraction and harder interpretation can use different settings.

The name Small does not make it a laptop model, and a European vendor does not establish European data residency. Choose the hosting location, retention policy and tool integrations that meet your requirements. This is a model choice; the deployment is a separate decision.

Muse Spark Open changes the Meta option

Muse Spark Open, from Meta, is the new Meta candidate for multimodal agents. Its 816B total / 42B active parameters and 1M-token context come with text, image, video and audio input and text output. For an agent that must understand a recording, inspect a screenshot and take a tool action, that is a broader input specification than a text-only coding model. Muse Spark Open card.

Meta names Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI as hosted inference partners. It names SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp and Unsloth for inference. The download and catalog routes include Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. A named host is an access option, not a guarantee of identical modalities or production availability. Meta announcement, deployment guide.

Self-hosting remains a substantial commitment. Meta's card lists at least 700GB of VRAM for NVFP4 and 2TB for BF16; its Apple route uses two M5 Ultra Mac Studios with MLX. Those are format-specific configurations, not a claim that an ordinary desktop can run the original weights. Model card and deployment requirements.

The Muse Model License permits commercial uses subject to its conditions. You must retain the required notices, comply with the Usage Policy and display Built with Muse Spark Open in one of the specified places. Section 3 requires notification and a separate written agreement when a licensee or affiliate provides the defined model-as-a-service offering and their aggregate gross revenue exceeds $20 million during any consecutive 12 months, before offering Muse or its derivatives as part of that business. It explicitly excludes internal use, research, other commercial activity that is not model as a service, and features of your own product when that product is not model as a service. This is not a blanket revenue cap on every application built with Muse. Muse Model License.

If you already run Llama 4 successfully, keep it as your baseline rather than migrating on release news alone. Its own community license still governs that checkpoint. For a new Meta evaluation, start with Muse and compare the actual task outcomes and deployment conditions. Llama 4 model card.

Who should pick what

The right model follows your binding constraint. Use these eight primary picks as a shortlist, with NVIDIA's vision model as an additional deployment candidate.

The cost math, honestly

Choose the model and the host separately. A downloadable checkpoint has no universal per-token price. A host's cache policy, peak schedule, supported inputs and capacity limits can change which offer works for your application. These rates were verified October 16, 2026; each line gives uncached input and output per million tokens unless stated otherwise.

• Z.ai, GLM-5.3: $1.40 input / $4.40 output, with cached input at $0.26. Official API pricing.
• DeepSeek: V4.1 Flash is $0.15 / $0.60 off-peak and $0.30 / $1.20 peak; Pro-0813 is $0.66 / $1.98 off-peak and $1.32 / $3.96 peak. Official rates and UTC schedule.
• Fireworks AI, Muse Spark Open: $1.25 input / $4.25 output, cached input $0.15; US-only serverless preview through October 29, 2026, with text/image input. Model pricing and preview terms.
• Baseten, Muse Spark Open: $1.25 input / $4.25 output, cached input $0.15; its model page lists text, image, video and audio inputs. Model API pricing.

The Muse headline rates match across those two hosts. Input support and capacity terms therefore deserve attention before price breaks the tie. For DeepSeek, price a workload against the peak schedule as well as the cheaper window; the old flat V4 rates are not the current offer.

1. 1

   Rent the API for the first evaluation

   Use representative tasks and record accepted outputs, latency, retries and billed tokens. The cheapest token rate is useful only if the model completes the work at the quality you require.

2. 2

   Check the subscription limits

   A coding subscription is a separate product. The GLM Coding Plan has both five-hour and weekly credit limits; it is not unlimited inference. Its current documentation also says older GLM-5.2 and GLM-5.1 plan requests route to GLM-5.3. Plan terms.

3. 3

   Self-host after sizing the complete workload

   Budget for the weights, request memory, concurrency, idle capacity and engineering time. Start with the model's documented format and serving recipe, then measure your workload. Active parameter count describes computation; it is not the amount of storage or memory you can assume you need.

There is no defensible universal savings multiplier here. Self-hosting may earn its cost through control or steady utilization; a hosted API may be cheaper for intermittent work. Compare completed jobs and the operating burden, not just the advertised token rate. For the wider coding choice, see the best AI model for coding.

The decision rule

Decide the deployment boundary first. If you can use an API, compare capability, host terms and the bill on your own tasks. If the model must run on infrastructure you control, filter by the exact license and a realistic hardware configuration before comparing benchmarks.

The next move is concrete: take a representative batch from your product, run your shortlisted model and current baseline with the same tools, and inspect the failures. Keep the model that meets your quality threshold within the deployment budget. A release earns a migration by improving that result.

What is the best open-source LLM in 2026?

There is no single winner across these deployment constraints. Evaluate GLM-5.3 for long coding workflows, DeepSeek V4.1 Flash for a low-cost hosted starting point, and gpt-oss-120b for local text reasoning. Muse Spark Open adds multimodal inputs and broad deployment support, with custom commercial terms.

Is open weight the same as open source?

No. Open weight describes access to trained parameters. The license determines permitted uses, and the Open Source AI Definition also requires code and training-data information. Check the exact release: a family's license can change between versions.

Can I run these models on my own hardware?

Yes, if the checkpoint's license and serving requirements fit your deployment. The released gpt-oss-120b quantization targets one 80GB GPU; the larger models need their own capacity plan. Muse's card describes substantially larger configurations, including its two-Mac-Studio route. Tool connections and logs still need their own data controls.

What is the best open LLM I can run locally on a normal machine?

gpt-oss-20b is the starting point here when you have the documented 16GB of memory and want text reasoning under Apache 2.0. Check the runtime and available memory before choosing it; the larger flagship models in this list are a different deployment class.

Want the full landscape mapped to your business, not just the models but the agents, tools, and stacks that turn them into products? Get the AI tools map for business owners and a short weekly read on what actually shipped and what to do about it.